Ilmu Komputer & AI editorial
From Hypervisor to Container: Cloud Security Vulnerabilities, Defense Mechanisms, and Open Challenges
The core problem
Cloud computing's economic model rests on multi-tenancy: different users share the same physical hardware, which creates serious security risks. To protect data, cloud systems rely on virtual machines (VMs) and containers to keep users isolated. The strength of these isolation boundaries determines the practical security of the entire cloud stack.
This paper reviews over 120 security publications from 2008 to 2025, focusing on how these isolation boundaries can be breached. The authors examine threats including virtual machine escape, virtual machine hopping, CPU cache side-channels, container breakouts, vulnerable container images, and distributed denial of service (DDoS) attacks. The review is organized around three key research questions that evaluate these security threats and their defenses.
A central contribution is the ADPO quantitative scoring framework, which rates defenses from 0 to 3 based on their **A**ccuracy, **D**eployment ease, **P**erformance impact, and **O**perational overhead. The paper also maps the impact of these attacks onto a 1-to-5 severity scale for Confidentiality, Integrity, and Availability (CIA). Finally, it highlights the trade-offs between securit
Innovation
The review identifies and evaluates a broad class of isolation-breaching threats across both VM and container layers. The main threat categories examined are:
1. **Virtual machine escape** — breaking out of a guest VM to the hypervisor or host.
2. **Virtual machine hopping** — moving laterally from one VM to another on the same host.
3. **CPU cache side-channels** — extracting sensitive data via shared microarchitectural state.
4. **Container breakouts** — escaping container isolation to the host or other containers.
5. **Vulnerable container images** — exploiting outdated or malicious image contents.
6. **Distributed denial of service (DDoS)** — exhausting shared resources to degrade availability.
Each threat is assessed using the three research questions and scored with the ADPO framework for defenses and the 1-to-5 CIA scale for impact. The results highlight that defenses vary widely in accuracy, deployability, performance cost, and operational overhead. Some defenses achieve high accuracy but impose significant performance penalties, while others are easy to deploy but offer limited protection.
The paper also reports that the security-performance trade-off is a recurring the
Why it matters
The paper's analysis centers on the trade-offs between security and system performance. Using the ADPO framework, defenses can be compared not only on how well they work (Accuracy) but also on how practical they are to deploy (Deployment ease), how much they slow the system (Performance impact), and how much ongoing effort they require (Operational overhead). This multidimensional view reveals that many high-accuracy defenses score poorly on performance and operational overhead, while lightweight defenses often sacrifice accuracy.
Mapping attack impact onto the 1-to-5 CIA severity scale shows that different threats prioritize different security goals. For example, CPU cache side-channels primarily threaten Confidentiality, DDoS attacks primarily threaten Availability, and container image vulnerabilities can threaten Integrity and Confidentiality. This helps practitioners prioritize defenses based on the CIA dimensions most critical to their workloads.
The review also highlights open challenges. First, building low-overhead intrusion detection remains difficult: effective detection often requires deep visibility that conflicts with performance goals. Second, creating realistic test datasets is a persistent problem, as real cloud traces are sensitive and hard to share, limiting reproducibility. Third, the rapid evolution of container ecosystems introduces new image and orchestration vulnerabilities faster than defenses can be standardized.
Overall, the study argues that cloud security is not a single boundary problem but a layered, continuously contested space. The ADPO and CIA scoring approaches provide a structured language for comparing defenses and impacts, but the fundamental security-performance trade-off means that no single solution is sufficient. Future work must focus on low-overhead, deployable defenses and on shared, realistic benchmarks to evaluate them.
Who should read this
Opening member content…